Unsupervised feature-preserving cyclegan for fault diagnosis of rolling bearings using unbalanced infrared thermal imaging / Lujiale Guo

Lujiale , Guo (2025) Unsupervised feature-preserving cyclegan for fault diagnosis of rolling bearings using unbalanced infrared thermal imaging / Lujiale Guo. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (117Kb)
	PDF (Thesis M.A) Download (1524Kb)

Abstract

The fault diagnosis of rolling bearing is of great significance in industrial safety. The method of infrared thermal image combined with neural network could diagnose the fault of rolling bearing in a non-contact manner, however its data in different scenes are often unbalanced and difficult to obtain. In this paper, an unsupervised learning framework named Feature-Preserving Cycle-Consistent Generative Adversarial Networks (FP-CycleGAN) is designed for defect detection in unbalanced rolling bearing infrared thermography sample. Since the classical Cycle-Consistent Generative Adversarial Networks (CycleGAN) not designed to accurately transfer the target features of the image. To avoid this problem, a new discriminator is designed to identify whether the generated image A and B (refer to different conditional bearing image) belongs to two different classes, and a new class loss are proposed. To better extract fault features and perform features migration, the new generator is reconstructed based on the U-Network structure, the transpose convolution method of the up-sampling network is replaced by Bicubic Interpolation to effectively avoid the checkerboard effect of the generated images. The defect detection of the expanded dataset was performed using Residual Network and compared with the pre-expansion data to demonstrate the usability of the generated data and the superiority of the proposed FP-CycleGAN method for rolling bearing defect detection in small samples of infrared thermal images. Finally, the accuracy of the proposed model is 91.52%, which is better than the baseline model (76.81%).

Actions (For repository staff only : Login required)